* 'master' of gitlab.physik.uni-muenchen.de:Nikolai.Hartmann/KerasROOTClassification:
  normalize weights for default training as well
  try normalising weights
  put model checkpoint/reload weights into separate function

* 'master' of gitlab.physik.uni-muenchen.de:Nikolai.Hartmann/KerasROOTClassification:
  (hopefully) correct treatment of class_weight
  little fix
  load and plot before training RNN
  fix write_friend_tree
  evaluate RNN working
  model and training for RNN working
  data structuring for RNN working
  remove validate and plot all from train
  put score evaluation in separate function
  correct quantile range for input plots
  making balanced training more efficient
  starting to develop yield_batch function for RNN
  starting rnn wrapper
  making balanced training more efficient

* 'master' of gitlab.physik.uni-muenchen.de:Nikolai.Hartmann/KerasROOTClassification:
  plot train and test for ROC by default
  commenting number of cores for session
  staring eval_model script
  allow both None and 0 for deactivating dropout
  try to avoid copying also when fitting scaler
  further memory improvements
  Memory improvements
  Experimental support for initialising from pandas DataFrame (memory intense ...)
  starting initialisation from pandas df
  support for tensorboard
  include identifiers in to_DataFrame
  adding to_DataFrame function
  Extended dropout support
  weighted accuracy
  weighted quantile for plot_NN_2D
  all neuron plot in plot_NN_2D script

* stop option for reading a maximum number of events for train/test
* inplace scaling